Wide-Baseline Stereo Vision for Mars Rovers
|
|
- Lorin Porter
- 5 years ago
- Views:
Transcription
1 Proceedings of the 2003 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems Las Vegas, Nevada October 2003 Wide-Baseline Stereo Vision for Mars Rovers Clark F. Olson Habib Abi-Rached Ming Ye Jonathan P. Hendrich University of Washington University of Washington Microsoft Corp. Computing and Software Systems Electrical Engineering th Ave. NE Box Box Redmond, WA Campus Way NE 253 EE/CSE Building Bothell, WA Seattle, WA Abstract In order to perform localization and navigation over significant distances (up to a few kilometers), it is important to be able to accurately map the terrain to be traversed. Local methods, such as conventional stereo, do not scale well to large distances and prevent longrange planning. In this paper, we discuss wide-baseline stereo techniques for rovers. In wide-baseline stereo, the image pair is captured with the same camera, but at different positions of the rover. While the larger baseline allows improved accuracy for more distant terrain, stereo matching is more difficult for two reasons. First, odometry errors result in uncertain knowledge of the relative positions of the cameras when the images are captured. Second, the change in perspective makes stereo matching difficult, since image landmarks no longer have the same appearance in both images. We address these problems and show test results on real images. 1 Introduction For the robotic exploration of Mars, a key goal is to maximize the amount of scientific data returned during the fixed span of a mission. This means that a rover must navigate accurately to science targets observed in orbital images or images captured while landing. Since communication with a rover on Mars is usually performed only once per day, the rover must be able to navigate to a target that it cannot see. If the rover fails to reach the goal, an entire day of scientific activity can be lost as the rover again attempts to reach the goal. Current work on rover mapping and localization has partially addressed this issue (for example, [3, 5, 7, 8]). One issue that has not, yet, been addressed is the computation by a rover of accurate maps of terrain many meters distant. This allows longer range planning and improved localization capabilities. Conventional stereo vision can generate accurate maps for close targets. However, the accuracy of these methods scales poorly for distant targets, since the range error increases with the square of the distance. One solution to this problem is to use a larger baseline (the distance between the cameras), which improves the accuracy of the range estimates. The obvious problem is that a rover with a limited size cannot have two cameras with a large baseline. We achieve an arbitrarily large baseline using two images captured by the rover at separate positions. While this can improve the accuracy of the range estimation, it introduces new problems. First, stereo algorithms typically calibrate a stereo pair of cameras such that the relative position and orientations of the cameras are known to high precision. This is not possible with wide-baseline stereo, since rover odometry errors prevent such high precision camera positioning. Second, the change in the viewpoint for the images makes stereo matching more difficult, since the terrain no longer has the same appearance in both images. Our algorithm addresses these problems and consists of the following steps: 1. Motion refinement. While our method assumes that an estimate of the camera positions is available, we do not assume that this estimate is accurate. We perform a motion refinement step in order to refine the relative camera position. The overall process consists of the following substeps: (a) Feature selection. Localizable features are first selected in one of the images using a /03/$ IEEE 1302
2 (c) Figure 1: Feature matching example. (a) Features selected in the left image. Initial features matched in right image. (c) Matched features in the left image after outlier rejection. (d) Matched features in the right image after outlier rejection. (d) variant of the Förstner interest operator [1]. Feature matching. Matches for the selected features are detected using a hierarchical search over the entire image. (c) Nonlinear motion estimation. The Levenberg-Marquardt method is used to optimize the motion parameters with respect to the image matches that have been detected. This optimization enforces the epipolar constraints in the stereo image pair. 2. Image rectification. In order to facilitate stereo matching, the images are rectified so that the epipolar lines lie along the scanlines of the image [2]. 3. Disparity calculation. Robust dense matching is performed using maximum-likelihood image matching techniques combined with efficient stereo search techniques. Subpixel disparities are computed by fitting the likelihood surface. 4. Triangulation. Once image disparities have been computed, triangulation is performed to determine the three-dimensional position of each of the image pixels. The remainder of this paper describes each of these steps and shows results using this methodology. 2 Feature selection and matching In order to refine the motion estimate between the two camera positions, we first detect interest points in one image and find the corresponding matches in the second image. The initial evaluation of each location in the first image is conducted using the Förstner interest operator [1]. This operator scores each pixel based on the strength of the gradients in a region around the pixel. For a pixel to score highly, it must have both strong local gradients in the image and isotropy of the gradient (so that linear edges are discarded.) This is accomplished by examining the covariance matrix of the local gradients. The largest eigenvalue of this matrix is an estimate of the strength of the local gradients and the ratio of the eigenvalues yields the degree of isotropy. To select individual features, we divide the image into sub-images and select the best local maxima in each sub-image. The reason for the subdivision of the image is to ensure that we select features in each part of the image. A match for each of the selected features 1303
3 is then sought in the second image. Matching between the images is performed using a hierarchical multi-resolution search for 7 7 templates centered at each of the selected features. Figure 1 shows an example of the features selected and matched using these techniques. In this case, 256 features were selected in the first image. Of these, candidate matches were found for 42 features in the second image, some of which are incorrect. After outlier rejection 22 correct matches remained for input to the motion refinement step. One direction of future work will be to increase the fraction of selected features that can be robustly matched, for example using the robust matching framework described below. 3 Motion refinement Motion refinement is the process of adjusting the initial estimate for the robot position and orientation using the stereo matches found in the image. Once the matches have been found as described in the previous section, we apply Levenberg-Marquardt optimization in order to refine the input motion estimate. In particular, we seek precise values for the translation T and rotation R relating the positions of the rover camera at the two locations. This is necessary so that the images can be rectified and the search space for the dense disparity estimation can be reduced to the corresponding scanline. Our optimization method is related to previous nonlinear methods for performing motion estimation from image data (see, for example, [9, 11]). Our state vector includes not only the six parameters describing the relative camera positions (only five are recoverable, since the problem can be scaled to an arbitrary size), but also the depth estimates to each of the recovered features. With this augmented state vector, the objective function that we are minimizing becomes the sum of squared distances between the detected feature position and the estimated feature position (calculated using backprojection from the current motion estimate): N i=1 ( (c i ĉ i ) 2 +(r i ˆr i ) 2), (1) where (r i,c i ) are the row and column position of the ith feature match in the second image and (ˆr i, ĉ i )are the predicted position of the ith feature in the second image using the current motion and depth estimate. The optimization is iterated, updating the motion and depth estimates at each step, until the objective function is minimized. The application of these techniques to the images in Figure 1 resulted in an average reprojection error per pixel (the values summed and squared in Eq. 1) of 0.34 pixels. In a series of similar tests, the average reprojection errors 0.31 pixels, with only two outliers having reprojection error greater than 1.0 pixel. 4 Image rectification In wide-baseline stereo matching, there is often a very large (positive or negative) disparity between matching points in the two images. This necessitates a large search space in the horizontal dimension for the correct match. If a search must also be performed in the vertical dimension, then the overall computation required will be large. We rectify the images using the refined motion estimate so that the matches will be along the corresponding row of the other image (assuming the refined motion estimate is accurate). This is achieved if the images are warped such that they appear as if each camera was rotated to have z-axis is perpendicular to the baseline and x-axis is parallel to the baseline. To accomplish the rectification, we use the method of Fusiello et al. [2]. Let R 1 and R 2 be the rotation matrices of the two cameras in the world coordinate frame and c 1 and c 2 be their centers of projection in this frame. The camera intrinsic parameters are the same for both images, since they are captured with the same camera. These intrinsic parameters are given by: A = α u γ u 0 0 α v v , (2) where α u and α v are the focal lengths in horizontal and vertical pixels, u 0 and v 0 are the image coordinates of the principal point, and γ is the skew factor (for non-orthogonal axes). A 3D point with coordinates [x yz] t is projected into an image point whose 2D image coordinates in the first image [u 1 v 1 ] t are given by the intersection of the image plane with the line containing the point and c 1. Using homogeneous coordinates, we can write: u 1 v 1 1 A[R 1 R 1 c 1 ] x y z 1 = P 1 x y z 1, (3) 1304
4 Figure 2: Images after rectification. Lines have been added to show the relative position of features in the two images. (a) Left image. Right image. where indicates proportionality. A similar relationship holds for coordinates in the second image. The rectifying transformation is a virtual rotation of the left and right camera around their respective centers of projection, with no change in translation. The new rotation R of both cameras in the world coordinate frame must be such that the new image planes are coplanar. This ensures that the epipoles are at infinity, and therefore the epipolar lines are parallel. To enforce horizontal epipolar lines, the baseline must be parallel to the x-axis in both cameras local frame of reference. The rotation is constructed of three row vectors as follows: R = r t 1 r t 2 r t 3 (4) 1. r 1 is a unit vector in the direction of c 1 c r 2 is a unit vector orthogonal to r 1 and to the z-axis in the local frame of reference of camera r 3 is a unit vector perpendicular to both r 1 and r 2. With the new rotation, the projection matrices become: P i = A[R Rc i ]=[Q i q i ] (5) and the rectifying transformation is: T i = Q i (AR i ) 1. (6) After applying the rectifying transformation, we recenter the images to capture as much of the data as possible within the same image boundaries. Figure 2 shows our sample images after rectifying them according to this process using the motion estimate provided by the previous optimization. The lines drawn in the images indicate that the rectification is good, with corresponding points lying on the same scanline. 5 Disparity estimation Given the rectified images, disparity estimation is performed by combining a robust template matching method [6] with an efficient stereo search strategy. Our experiments have indicated that the use of a matching measure such as the SSD produces very poor results. We use a maximum-likelihood measure that uses distance measurements from each pixel in the image template to the closest corresponding pixel in the search image. D i (δ) is the distance for the ith template pixel at disparity δ. Assuming the distances are independent, our likelihood function is: L(D 1 (δ),..., D m (δ) δ) = m p(d i (δ)), (7) i=1 where p(d i (δ)) is the probability density function (PDF) of D i (δ) evaluated at the template disparity δ. In order to find the correct disparity, we find the displacement δ that maximizes the above likelihood function. Our maximum-likelihood measure improves upon the SSD in two important ways. First, the SSD measure compares only the pixels that are directly overlapping at some disparity of the template with respect to the search image. If camera motion or perspective distortion causes pixels to move by different amounts between the two images, then it will not be possible to find a template position where all of the pixels are correctly overlapped. Our distance measure allows pixels that are not directly overlapping to be matched by linearly combining the distance in the image with the difference in intensity. Computing the best distance for a pixel at a particular template position is no longer trivial with this formulation. However, efficient computation of the distances can be performed by precomputing a three-dimensional distance transform of the input data [6]. 1305
5 Figure 3: Computed disparity map. (a) Right image. Disparity image. Dark values represent larger disparities. The other improvement over SSD matching is that the possibility of an outlier is explicitly represented. In this application, any terrain feature that is visible in one of the images, but not in the other is an outlier for the matching process. Such outliers occur frequently for images taken from different viewpoints. In order to model such outliers, we use a probability density function for each pixel distance that has two terms, one for inliers and one for outliers: p(d) =αp 1 (d)+(1 α)p 2 (d). (8) The first term is the error density for inliers and the second term is the error density for outliers, where α is the probability that a particular distance represents an inlier. For inliers, we use a normal distribution in the distance to the closest pixel. For outliers, we use a constant. (This is not a true probability density, but it allows us to model each distance as equally likely in the case of an outlier.) In order to perform dense matching between the rectified images using the measure described above, we use an efficient search strategy common in stereo vision. This strategy makes use of the observation that a brute-force implementation performs many redundant computations for adjacent positions of the template at the same displacement. We eliminate the redundant computation by storing the information for reuse as necessary for fast matching. Once the integral disparity estimates have been computed using template matching, we compute a subpixel disparity estimate and an estimate of the standard deviation of the error for each pixel. This is performed by fitting a curve to the likelihood scores near the maximum [5]. Disparity estimates are pruned if the expected error is too large or if the likelihood of the selected location is not large enough. Figure 3 shows the disparity map that was computed for the example in the previous figures. In this case, we obtain sparse results in some portions of the image. This occurs for several reasons. First, there is insufficient texture in thy sky to correctly match the pixels. Second, there are large regions of the image that are not visible in the other image. Note how the high quality disparity matches for the mountains in the background end abruptly about a third of an image width from the right hand side. This is because the mountains to the left of this region are not present in the other image. Finally, the significant change in the viewpoint (approximately 20 degrees) makes matching difficult, particularly in the foreground, where there is very little similarity between the correctly matching features. Figure 4 shows a second example. Denser results are obtained in this case, since the cameras were pointing in nearly the same directly and the baseline was considerably smaller. Data is correctly pruned from the lower right portion of the image, since this region is not visible in the other image. 6 Rover integration The wide-baseline stereo method has been integrated with the CLARAty architecture for robotic autonomy [4, 10], which is designed for modularity using object-oriented methods. CLARAty uses generic functional classes that specify the interface and functionality of a generic algorithm. Stereo vision is an example of such a generic functional class. The stereo vision class uses generic image and camera classes in its implementation. In turn, the stereo vision class can be used in the implementation of a visual navigation class. Our wide-baseline stereo technique has been implemented as specialization of the generic stereo vision class. This allows the wide-baseline stereo techniques 1306
6 Figure 4: Computed disparity map. (a) Right image. Disparity image. Dark values represent larger disparities. to be used in any place that the generic stereo vision class is used in the CLARAty architecture. The current computation time required for the code is around 120 seconds on a 500 MHz workstation. We believe that this can be improved and that a significant computation time is reasonable, since this operation would be performed infrequently. 7 Summary We have described techniques to allow widebaseline stereo to be performed on a Mars rover. With these techniques, the rover can accurately map terrain many meters away, unlike conventional stereo. We have overcome two significant problems to achieve good results. Inexact knowledge of the relative positions of the cameras has been addressed using a nonlinear motion estimation step. This step automatically selects and matches features in the images in order to refine the initial motion estimate. The problem of robust matching between terrain viewed from different perspectives has been addressed using a robust template matching method that tolerates outliers and perspective distortion. The overall method has achieved good stereo matching results even with a signification change in image viewpoint. Acknowledgments We gratefully acknowledge funding of this work by the NASA Mars Technology Program. References Vision, pages , [2] A. Fusiello, E. Trucco, and A. Verri. A compact algorithm for rectification of stereo pairs. Machine Vision and Applications, 12:16 22, [3] R. Li, F. Ma, L. H. Matthies, C. F. Olson, and R. E. Arvidson. Localization of Mars rovers using descent and surface-based image data. Journal of Geophysical Research -Planets, [4] I. A. D. Nesnas, R. Volpe, T. Estlin, H. Das, R. Petras, and D. Mutz. Toward developing reusable software components for robotic applications. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, volume 4, pages , [5] C. F. Olson. Probabilistic self-localization for mobile robots. IEEE Transactions on Robotics and Automation, 16(1):55 66, February [6] C. F. Olson. Maximum-likelihood image matching. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(6): , June [7] C.F.Olson,L.H.Matthies,M.Schoppers,andM.W. Maimone. Robust stereo ego-motion for long distance navigation. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, volume 2, pages , [8] C. F. Olson, L. H. Matthies, Y. Xiong, R. Li, F. Ma, and F. Xu. Multi-resolution mapping using surface, descent, and orbital images. In Proceedings of the 6th International Symposium on Artificial Intelligence, Robotics and Automation in Space, [9] R. Szeliski and S. B. Kang. Recovering 3d shape and motion from image streams using non-linear least squares. Journal of Visual Communication and Image Representation, 5(1):10 28, March [10] R. Volpe, I. Nesnas, T. Estlin, D. Mutz, R. Petras, and H. Das. The CLARAty architecture for robotic autonomy. In Proceedings of the 2001 IEEE Aerospace Conference, volume 1, pages , [11] Y. Xiong, C. F. Olson, and L. H. Matthies. Computing depth maps from descent imagery. In Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, volume 1, pages , [1] W. Förstner. A framework for low-level feature extraction. In Proceedings of the European Conference on Computer 1307
Step-by-Step Model Buidling
Step-by-Step Model Buidling Review Feature selection Feature selection Feature correspondence Camera Calibration Euclidean Reconstruction Landing Augmented Reality Vision Based Control Sparse Structure
More informationRectification and Distortion Correction
Rectification and Distortion Correction Hagen Spies March 12, 2003 Computer Vision Laboratory Department of Electrical Engineering Linköping University, Sweden Contents Distortion Correction Rectification
More informationCorrespondence and Stereopsis. Original notes by W. Correa. Figures from [Forsyth & Ponce] and [Trucco & Verri]
Correspondence and Stereopsis Original notes by W. Correa. Figures from [Forsyth & Ponce] and [Trucco & Verri] Introduction Disparity: Informally: difference between two pictures Allows us to gain a strong
More informationRectification and Disparity
Rectification and Disparity Nassir Navab Slides prepared by Christian Unger What is Stereo Vision? Introduction A technique aimed at inferring dense depth measurements efficiently using two cameras. Wide
More informationStereo Image Rectification for Simple Panoramic Image Generation
Stereo Image Rectification for Simple Panoramic Image Generation Yun-Suk Kang and Yo-Sung Ho Gwangju Institute of Science and Technology (GIST) 261 Cheomdan-gwagiro, Buk-gu, Gwangju 500-712 Korea Email:{yunsuk,
More informationStereo Vision. MAN-522 Computer Vision
Stereo Vision MAN-522 Computer Vision What is the goal of stereo vision? The recovery of the 3D structure of a scene using two or more images of the 3D scene, each acquired from a different viewpoint in
More informationStereo and Epipolar geometry
Previously Image Primitives (feature points, lines, contours) Today: Stereo and Epipolar geometry How to match primitives between two (multiple) views) Goals: 3D reconstruction, recognition Jana Kosecka
More information3D Sensing and Reconstruction Readings: Ch 12: , Ch 13: ,
3D Sensing and Reconstruction Readings: Ch 12: 12.5-6, Ch 13: 13.1-3, 13.9.4 Perspective Geometry Camera Model Stereo Triangulation 3D Reconstruction by Space Carving 3D Shape from X means getting 3D coordinates
More information55:148 Digital Image Processing Chapter 11 3D Vision, Geometry
55:148 Digital Image Processing Chapter 11 3D Vision, Geometry Topics: Basics of projective geometry Points and hyperplanes in projective space Homography Estimating homography from point correspondence
More informationImage Rectification (Stereo) (New book: 7.2.1, old book: 11.1)
Image Rectification (Stereo) (New book: 7.2.1, old book: 11.1) Guido Gerig CS 6320 Spring 2013 Credits: Prof. Mubarak Shah, Course notes modified from: http://www.cs.ucf.edu/courses/cap6411/cap5415/, Lecture
More informationPrecise Omnidirectional Camera Calibration
Precise Omnidirectional Camera Calibration Dennis Strelow, Jeffrey Mishler, David Koes, and Sanjiv Singh Carnegie Mellon University {dstrelow, jmishler, dkoes, ssingh}@cs.cmu.edu Abstract Recent omnidirectional
More informationMultiview Stereo COSC450. Lecture 8
Multiview Stereo COSC450 Lecture 8 Stereo Vision So Far Stereo and epipolar geometry Fundamental matrix captures geometry 8-point algorithm Essential matrix with calibrated cameras 5-point algorithm Intersect
More informationQuasi-Euclidean Uncalibrated Epipolar Rectification
Dipartimento di Informatica Università degli Studi di Verona Rapporto di ricerca Research report September 2006 RR 43/2006 Quasi-Euclidean Uncalibrated Epipolar Rectification L. Irsara A. Fusiello Questo
More informationVisual Odometry. Features, Tracking, Essential Matrix, and RANSAC. Stephan Weiss Computer Vision Group NASA-JPL / CalTech
Visual Odometry Features, Tracking, Essential Matrix, and RANSAC Stephan Weiss Computer Vision Group NASA-JPL / CalTech Stephan.Weiss@ieee.org (c) 2013. Government sponsorship acknowledged. Outline The
More informationCOMPARATIVE STUDY OF DIFFERENT APPROACHES FOR EFFICIENT RECTIFICATION UNDER GENERAL MOTION
COMPARATIVE STUDY OF DIFFERENT APPROACHES FOR EFFICIENT RECTIFICATION UNDER GENERAL MOTION Mr.V.SRINIVASA RAO 1 Prof.A.SATYA KALYAN 2 DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING PRASAD V POTLURI SIDDHARTHA
More informationStereo II CSE 576. Ali Farhadi. Several slides from Larry Zitnick and Steve Seitz
Stereo II CSE 576 Ali Farhadi Several slides from Larry Zitnick and Steve Seitz Camera parameters A camera is described by several parameters Translation T of the optical center from the origin of world
More informationA Specialized Multibaseline Stereo Technique for Obstacle Detection
A Specialized Multibaseline Stereo Technique for Obstacle Detection Todd Williamson and Charles Thorpe Robotics Institute Carnegie Mellon University Pittsburgh PA 15213 Abstract This paper presents a multibaseline
More informationMaximum-Likelihood Image Matching
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 24, NO. 6, JUNE 2002 853 Maximum-Likelihood Image Matching Clark F. Olson, Member, IEEE AbstractÐImage matching applications such as
More informationComputer Vision Lecture 17
Computer Vision Lecture 17 Epipolar Geometry & Stereo Basics 13.01.2015 Bastian Leibe RWTH Aachen http://www.vision.rwth-aachen.de leibe@vision.rwth-aachen.de Announcements Seminar in the summer semester
More informationComputer Vision Lecture 17
Announcements Computer Vision Lecture 17 Epipolar Geometry & Stereo Basics Seminar in the summer semester Current Topics in Computer Vision and Machine Learning Block seminar, presentations in 1 st week
More informationWide-baseline stereo vision for terrain mapping
Machine Vision and Applications (00) :73 75 DOI 0.007/s0038-009-088-9 ORIGINAL PAPER Wide-baseline stereo vision for terrain mapping Clark F. Olson Habib Abi-Rached Received: January 008 / Revised: 30
More informationApplication questions. Theoretical questions
The oral exam will last 30 minutes and will consist of one application question followed by two theoretical questions. Please find below a non exhaustive list of possible application questions. The list
More informationLecture 14: Basic Multi-View Geometry
Lecture 14: Basic Multi-View Geometry Stereo If I needed to find out how far point is away from me, I could use triangulation and two views scene point image plane optical center (Graphic from Khurram
More informationStereo CSE 576. Ali Farhadi. Several slides from Larry Zitnick and Steve Seitz
Stereo CSE 576 Ali Farhadi Several slides from Larry Zitnick and Steve Seitz Why do we perceive depth? What do humans use as depth cues? Motion Convergence When watching an object close to us, our eyes
More information1 Projective Geometry
CIS8, Machine Perception Review Problem - SPRING 26 Instructions. All coordinate systems are right handed. Projective Geometry Figure : Facade rectification. I took an image of a rectangular object, and
More informationFlexible Calibration of a Portable Structured Light System through Surface Plane
Vol. 34, No. 11 ACTA AUTOMATICA SINICA November, 2008 Flexible Calibration of a Portable Structured Light System through Surface Plane GAO Wei 1 WANG Liang 1 HU Zhan-Yi 1 Abstract For a portable structured
More informationVisual Terrain Mapping for Mars Exploration 1,2
Visual Terrain Mapping for Mars Exploration 1,2 Clark F. Olson Computing and Software Systems University of Washington, Bothell 18115 Campus Way NE, Box 358534 Bothell, WA 98011-8246 +1-425-352-5288 cfolson@u.washington.edu
More informationUsing temporal seeding to constrain the disparity search range in stereo matching
Using temporal seeding to constrain the disparity search range in stereo matching Thulani Ndhlovu Mobile Intelligent Autonomous Systems CSIR South Africa Email: tndhlovu@csir.co.za Fred Nicolls Department
More informationCHAPTER 3 DISPARITY AND DEPTH MAP COMPUTATION
CHAPTER 3 DISPARITY AND DEPTH MAP COMPUTATION In this chapter we will discuss the process of disparity computation. It plays an important role in our caricature system because all 3D coordinates of nodes
More informationIndex. 3D reconstruction, point algorithm, point algorithm, point algorithm, point algorithm, 263
Index 3D reconstruction, 125 5+1-point algorithm, 284 5-point algorithm, 270 7-point algorithm, 265 8-point algorithm, 263 affine point, 45 affine transformation, 57 affine transformation group, 57 affine
More informationPropagating these values through the probability density function yields a bound on the likelihood score that can be achieved by any position in the c
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2000. Maximum-Likelihood Template Matching Clark F. Olson Jet Propulsion Laboratory, California Institute of Technology 4800
More informationStereo imaging ideal geometry
Stereo imaging ideal geometry (X,Y,Z) Z f (x L,y L ) f (x R,y R ) Optical axes are parallel Optical axes separated by baseline, b. Line connecting lens centers is perpendicular to the optical axis, and
More informationCS5670: Computer Vision
CS5670: Computer Vision Noah Snavely, Zhengqi Li Stereo Single image stereogram, by Niklas Een Mark Twain at Pool Table", no date, UCR Museum of Photography Stereo Given two images from different viewpoints
More informationToday. Stereo (two view) reconstruction. Multiview geometry. Today. Multiview geometry. Computational Photography
Computational Photography Matthias Zwicker University of Bern Fall 2009 Today From 2D to 3D using multiple views Introduction Geometry of two views Stereo matching Other applications Multiview geometry
More informationIndex. 3D reconstruction, point algorithm, point algorithm, point algorithm, point algorithm, 253
Index 3D reconstruction, 123 5+1-point algorithm, 274 5-point algorithm, 260 7-point algorithm, 255 8-point algorithm, 253 affine point, 43 affine transformation, 55 affine transformation group, 55 affine
More informationEpipolar Geometry and Stereo Vision
Epipolar Geometry and Stereo Vision Computer Vision Jia-Bin Huang, Virginia Tech Many slides from S. Seitz and D. Hoiem Last class: Image Stitching Two images with rotation/zoom but no translation. X x
More informationCamera Calibration. Schedule. Jesus J Caban. Note: You have until next Monday to let me know. ! Today:! Camera calibration
Camera Calibration Jesus J Caban Schedule! Today:! Camera calibration! Wednesday:! Lecture: Motion & Optical Flow! Monday:! Lecture: Medical Imaging! Final presentations:! Nov 29 th : W. Griffin! Dec 1
More informationReminder: Lecture 20: The Eight-Point Algorithm. Essential/Fundamental Matrix. E/F Matrix Summary. Computing F. Computing F from Point Matches
Reminder: Lecture 20: The Eight-Point Algorithm F = -0.00310695-0.0025646 2.96584-0.028094-0.00771621 56.3813 13.1905-29.2007-9999.79 Readings T&V 7.3 and 7.4 Essential/Fundamental Matrix E/F Matrix Summary
More informationDense 3D Reconstruction. Christiano Gava
Dense 3D Reconstruction Christiano Gava christiano.gava@dfki.de Outline Previous lecture: structure and motion II Structure and motion loop Triangulation Today: dense 3D reconstruction The matching problem
More informationCamera Geometry II. COS 429 Princeton University
Camera Geometry II COS 429 Princeton University Outline Projective geometry Vanishing points Application: camera calibration Application: single-view metrology Epipolar geometry Application: stereo correspondence
More informationDepth. Common Classification Tasks. Example: AlexNet. Another Example: Inception. Another Example: Inception. Depth
Common Classification Tasks Recognition of individual objects/faces Analyze object-specific features (e.g., key points) Train with images from different viewing angles Recognition of object classes Analyze
More informationcse 252c Fall 2004 Project Report: A Model of Perpendicular Texture for Determining Surface Geometry
cse 252c Fall 2004 Project Report: A Model of Perpendicular Texture for Determining Surface Geometry Steven Scher December 2, 2004 Steven Scher SteveScher@alumni.princeton.edu Abstract Three-dimensional
More informationEfficient Stereo Image Rectification Method Using Horizontal Baseline
Efficient Stereo Image Rectification Method Using Horizontal Baseline Yun-Suk Kang and Yo-Sung Ho School of Information and Communicatitions Gwangju Institute of Science and Technology (GIST) 261 Cheomdan-gwagiro,
More information55:148 Digital Image Processing Chapter 11 3D Vision, Geometry
55:148 Digital Image Processing Chapter 11 3D Vision, Geometry Topics: Basics of projective geometry Points and hyperplanes in projective space Homography Estimating homography from point correspondence
More informationDense 3D Reconstruction. Christiano Gava
Dense 3D Reconstruction Christiano Gava christiano.gava@dfki.de Outline Previous lecture: structure and motion II Structure and motion loop Triangulation Wide baseline matching (SIFT) Today: dense 3D reconstruction
More informationcalibrated coordinates Linear transformation pixel coordinates
1 calibrated coordinates Linear transformation pixel coordinates 2 Calibration with a rig Uncalibrated epipolar geometry Ambiguities in image formation Stratified reconstruction Autocalibration with partial
More informationImage Transfer Methods. Satya Prakash Mallick Jan 28 th, 2003
Image Transfer Methods Satya Prakash Mallick Jan 28 th, 2003 Objective Given two or more images of the same scene, the objective is to synthesize a novel view of the scene from a view point where there
More informationModel-Based Stereo. Chapter Motivation. The modeling system described in Chapter 5 allows the user to create a basic model of a
96 Chapter 7 Model-Based Stereo 7.1 Motivation The modeling system described in Chapter 5 allows the user to create a basic model of a scene, but in general the scene will have additional geometric detail
More informationLong-term motion estimation from images
Long-term motion estimation from images Dennis Strelow 1 and Sanjiv Singh 2 1 Google, Mountain View, CA, strelow@google.com 2 Carnegie Mellon University, Pittsburgh, PA, ssingh@cmu.edu Summary. Cameras
More informationBIL Computer Vision Apr 16, 2014
BIL 719 - Computer Vision Apr 16, 2014 Binocular Stereo (cont d.), Structure from Motion Aykut Erdem Dept. of Computer Engineering Hacettepe University Slide credit: S. Lazebnik Basic stereo matching algorithm
More informationUnit 3 Multiple View Geometry
Unit 3 Multiple View Geometry Relations between images of a scene Recovering the cameras Recovering the scene structure http://www.robots.ox.ac.uk/~vgg/hzbook/hzbook1.html 3D structure from images Recover
More informationComputer Vision I. Announcement. Stereo Vision Outline. Stereo II. CSE252A Lecture 15
Announcement Stereo II CSE252A Lecture 15 HW3 assigned No class on Thursday 12/6 Extra class on Tuesday 12/4 at 6:30PM in WLH Room 2112 Mars Exploratory Rovers: Spirit and Opportunity Stereo Vision Outline
More informationModel Based Perspective Inversion
Model Based Perspective Inversion A. D. Worrall, K. D. Baker & G. D. Sullivan Intelligent Systems Group, Department of Computer Science, University of Reading, RG6 2AX, UK. Anthony.Worrall@reading.ac.uk
More informationAdaptive Zoom Distance Measuring System of Camera Based on the Ranging of Binocular Vision
Adaptive Zoom Distance Measuring System of Camera Based on the Ranging of Binocular Vision Zhiyan Zhang 1, Wei Qian 1, Lei Pan 1 & Yanjun Li 1 1 University of Shanghai for Science and Technology, China
More information3D Sensing. 3D Shape from X. Perspective Geometry. Camera Model. Camera Calibration. General Stereo Triangulation.
3D Sensing 3D Shape from X Perspective Geometry Camera Model Camera Calibration General Stereo Triangulation 3D Reconstruction 3D Shape from X shading silhouette texture stereo light striping motion mainly
More informationHomographies and RANSAC
Homographies and RANSAC Computer vision 6.869 Bill Freeman and Antonio Torralba March 30, 2011 Homographies and RANSAC Homographies RANSAC Building panoramas Phototourism 2 Depth-based ambiguity of position
More informationHIGH PRECISION LANDING SITE MAPPING AND ROVER LOCALIZATION BY INTEGRATED BUNDLE ADJUSTMENT OF MPF SURFACE IMAGES
ISPRS SIPT IGU UCI CIG ACSG Table of contents Table des matières Authors index Index des auteurs Search Recherches Exit Sortir HIGH PRECISION LANDING SITE MAPPING AND ROVER LOCALIZATION BY INTEGRATED BUNDLE
More informationStereo Vision Computer Vision (Kris Kitani) Carnegie Mellon University
Stereo Vision 16-385 Computer Vision (Kris Kitani) Carnegie Mellon University What s different between these two images? Objects that are close move more or less? The amount of horizontal movement is
More informationStereo. 11/02/2012 CS129, Brown James Hays. Slides by Kristen Grauman
Stereo 11/02/2012 CS129, Brown James Hays Slides by Kristen Grauman Multiple views Multi-view geometry, matching, invariant features, stereo vision Lowe Hartley and Zisserman Why multiple views? Structure
More informationA Compact Algorithm for Rectification of Stereo Pairs
Machine Vision and Applications manuscript No. (will be inserted by the editor) A Compact Algorithm for Rectification of Stereo Pairs Andrea Fusiello Emanuele Trucco Alessandro Verri Received: 25 February
More informationA Robust Two Feature Points Based Depth Estimation Method 1)
Vol.31, No.5 ACTA AUTOMATICA SINICA September, 2005 A Robust Two Feature Points Based Depth Estimation Method 1) ZHONG Zhi-Guang YI Jian-Qiang ZHAO Dong-Bin (Laboratory of Complex Systems and Intelligence
More informationSegmentation and Tracking of Partial Planar Templates
Segmentation and Tracking of Partial Planar Templates Abdelsalam Masoud William Hoff Colorado School of Mines Colorado School of Mines Golden, CO 800 Golden, CO 800 amasoud@mines.edu whoff@mines.edu Abstract
More informationIMAGE-BASED 3D ACQUISITION TOOL FOR ARCHITECTURAL CONSERVATION
IMAGE-BASED 3D ACQUISITION TOOL FOR ARCHITECTURAL CONSERVATION Joris Schouteden, Marc Pollefeys, Maarten Vergauwen, Luc Van Gool Center for Processing of Speech and Images, K.U.Leuven, Kasteelpark Arenberg
More informationHigh Accuracy Depth Measurement using Multi-view Stereo
High Accuracy Depth Measurement using Multi-view Stereo Trina D. Russ and Anthony P. Reeves School of Electrical Engineering Cornell University Ithaca, New York 14850 tdr3@cornell.edu Abstract A novel
More informationAnnouncements. Stereo
Announcements Stereo Homework 2 is due today, 11:59 PM Homework 3 will be assigned today Reading: Chapter 7: Stereopsis CSE 152 Lecture 8 Binocular Stereopsis: Mars Given two images of a scene where relative
More information3D Geometry and Camera Calibration
3D Geometry and Camera Calibration 3D Coordinate Systems Right-handed vs. left-handed x x y z z y 2D Coordinate Systems 3D Geometry Basics y axis up vs. y axis down Origin at center vs. corner Will often
More informationEE795: Computer Vision and Intelligent Systems
EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 FDH 204 Lecture 14 130307 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Stereo Dense Motion Estimation Translational
More informationWhat have we leaned so far?
What have we leaned so far? Camera structure Eye structure Project 1: High Dynamic Range Imaging What have we learned so far? Image Filtering Image Warping Camera Projection Model Project 2: Panoramic
More informationImage correspondences and structure from motion
Image correspondences and structure from motion http://graphics.cs.cmu.edu/courses/15-463 15-463, 15-663, 15-862 Computational Photography Fall 2017, Lecture 20 Course announcements Homework 5 posted.
More informationCS 664 Slides #9 Multi-Camera Geometry. Prof. Dan Huttenlocher Fall 2003
CS 664 Slides #9 Multi-Camera Geometry Prof. Dan Huttenlocher Fall 2003 Pinhole Camera Geometric model of camera projection Image plane I, which rays intersect Camera center C, through which all rays pass
More informationCS 4495 Computer Vision A. Bobick. Motion and Optic Flow. Stereo Matching
Stereo Matching Fundamental matrix Let p be a point in left image, p in right image l l Epipolar relation p maps to epipolar line l p maps to epipolar line l p p Epipolar mapping described by a 3x3 matrix
More informationTowards the completion of assignment 1
Towards the completion of assignment 1 What to do for calibration What to do for point matching What to do for tracking What to do for GUI COMPSCI 773 Feature Point Detection Why study feature point detection?
More informationThere are many cues in monocular vision which suggests that vision in stereo starts very early from two similar 2D images. Lets see a few...
STEREO VISION The slides are from several sources through James Hays (Brown); Srinivasa Narasimhan (CMU); Silvio Savarese (U. of Michigan); Bill Freeman and Antonio Torralba (MIT), including their own
More informationEpipolar Geometry and Stereo Vision
Epipolar Geometry and Stereo Vision Computer Vision Shiv Ram Dubey, IIIT Sri City Many slides from S. Seitz and D. Hoiem Last class: Image Stitching Two images with rotation/zoom but no translation. X
More informationEE795: Computer Vision and Intelligent Systems
EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 FDH 204 Lecture 12 130228 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Panoramas, Mosaics, Stitching Two View Geometry
More informationCSE 252B: Computer Vision II
CSE 252B: Computer Vision II Lecturer: Serge Belongie Scribes: Jeremy Pollock and Neil Alldrin LECTURE 14 Robust Feature Matching 14.1. Introduction Last lecture we learned how to find interest points
More informationAccurate and Dense Wide-Baseline Stereo Matching Using SW-POC
Accurate and Dense Wide-Baseline Stereo Matching Using SW-POC Shuji Sakai, Koichi Ito, Takafumi Aoki Graduate School of Information Sciences, Tohoku University, Sendai, 980 8579, Japan Email: sakai@aoki.ecei.tohoku.ac.jp
More informationLecture 9: Epipolar Geometry
Lecture 9: Epipolar Geometry Professor Fei Fei Li Stanford Vision Lab 1 What we will learn today? Why is stereo useful? Epipolar constraints Essential and fundamental matrix Estimating F (Problem Set 2
More informationMosaics. Today s Readings
Mosaics VR Seattle: http://www.vrseattle.com/ Full screen panoramas (cubic): http://www.panoramas.dk/ Mars: http://www.panoramas.dk/fullscreen3/f2_mars97.html Today s Readings Szeliski and Shum paper (sections
More informationLecture 6 Stereo Systems Multi-view geometry
Lecture 6 Stereo Systems Multi-view geometry Professor Silvio Savarese Computational Vision and Geometry Lab Silvio Savarese Lecture 6-5-Feb-4 Lecture 6 Stereo Systems Multi-view geometry Stereo systems
More informationA Novel Stereo Camera System by a Biprism
528 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 16, NO. 5, OCTOBER 2000 A Novel Stereo Camera System by a Biprism DooHyun Lee and InSo Kweon, Member, IEEE Abstract In this paper, we propose a novel
More information1 CSE 252A Computer Vision I Fall 2017
Assignment 1 CSE A Computer Vision I Fall 01 1.1 Assignment This assignment contains theoretical and programming exercises. If you plan to submit hand written answers for theoretical exercises, please
More informationStereo vision. Many slides adapted from Steve Seitz
Stereo vision Many slides adapted from Steve Seitz What is stereo vision? Generic problem formulation: given several images of the same object or scene, compute a representation of its 3D shape What is
More informationStructure from motion
Structure from motion Structure from motion Given a set of corresponding points in two or more images, compute the camera parameters and the 3D point coordinates?? R 1,t 1 R 2,t R 2 3,t 3 Camera 1 Camera
More informationCS 532: 3D Computer Vision 7 th Set of Notes
1 CS 532: 3D Computer Vision 7 th Set of Notes Instructor: Philippos Mordohai Webpage: www.cs.stevens.edu/~mordohai E-mail: Philippos.Mordohai@stevens.edu Office: Lieb 215 Logistics No class on October
More informationA Real-Time Catadioptric Stereo System Using Planar Mirrors
A Real-Time Catadioptric Stereo System Using Planar Mirrors Joshua Gluckman Shree K. Nayar Department of Computer Science Columbia University New York, NY 10027 Abstract By using mirror reflections of
More information3D Model Acquisition by Tracking 2D Wireframes
3D Model Acquisition by Tracking 2D Wireframes M. Brown, T. Drummond and R. Cipolla {96mab twd20 cipolla}@eng.cam.ac.uk Department of Engineering University of Cambridge Cambridge CB2 1PZ, UK Abstract
More informationCamera Calibration. COS 429 Princeton University
Camera Calibration COS 429 Princeton University Point Correspondences What can you figure out from point correspondences? Noah Snavely Point Correspondences X 1 X 4 X 3 X 2 X 5 X 6 X 7 p 1,1 p 1,2 p 1,3
More informationEXAM SOLUTIONS. Image Processing and Computer Vision Course 2D1421 Monday, 13 th of March 2006,
School of Computer Science and Communication, KTH Danica Kragic EXAM SOLUTIONS Image Processing and Computer Vision Course 2D1421 Monday, 13 th of March 2006, 14.00 19.00 Grade table 0-25 U 26-35 3 36-45
More informationCS6670: Computer Vision
CS6670: Computer Vision Noah Snavely Lecture 7: Image Alignment and Panoramas What s inside your fridge? http://www.cs.washington.edu/education/courses/cse590ss/01wi/ Projection matrix intrinsics projection
More informationIntroduction to Computer Vision
Introduction to Computer Vision Michael J. Black Nov 2009 Perspective projection and affine motion Goals Today Perspective projection 3D motion Wed Projects Friday Regularization and robust statistics
More informationRecap: Features and filters. Recap: Grouping & fitting. Now: Multiple views 10/29/2008. Epipolar geometry & stereo vision. Why multiple views?
Recap: Features and filters Epipolar geometry & stereo vision Tuesday, Oct 21 Kristen Grauman UT-Austin Transforming and describing images; textures, colors, edges Recap: Grouping & fitting Now: Multiple
More informationProf. Fanny Ficuciello Robotics for Bioengineering Visual Servoing
Visual servoing vision allows a robotic system to obtain geometrical and qualitative information on the surrounding environment high level control motion planning (look-and-move visual grasping) low level
More informationCamera Model and Calibration. Lecture-12
Camera Model and Calibration Lecture-12 Camera Calibration Determine extrinsic and intrinsic parameters of camera Extrinsic 3D location and orientation of camera Intrinsic Focal length The size of the
More informationStructure from motion
Structure from motion Structure from motion Given a set of corresponding points in two or more images, compute the camera parameters and the 3D point coordinates?? R 1,t 1 R 2,t 2 R 3,t 3 Camera 1 Camera
More informationClark F. Olson. Jet Propulsion Laboratory, California Institute of Technology Oak Grove Drive, MS , Pasadena, CA 91109
Proceedings of the 8th International Conference on Advanced Robotics, pages 447-452, 1997. Mobile Robot Self-Localization by Iconic Matching of Range Maps Clark F. Olson Jet Propulsion Laboratory, California
More informationRectification. Dr. Gerhard Roth
Rectification Dr. Gerhard Roth Problem Definition Given a pair of stereo images, the intrinsic parameters of each camera, and the extrinsic parameters of the system, R, and, compute the image transformation
More informationEpipolar geometry. x x
Two-view geometry Epipolar geometry X x x Baseline line connecting the two camera centers Epipolar Plane plane containing baseline (1D family) Epipoles = intersections of baseline with image planes = projections
More informationAnnouncements. Stereo
Announcements Stereo Homework 1 is due today, 11:59 PM Homework 2 will be assigned on Thursday Reading: Chapter 7: Stereopsis CSE 252A Lecture 8 Binocular Stereopsis: Mars Given two images of a scene where
More informationTwo-view geometry Computer Vision Spring 2018, Lecture 10
Two-view geometry http://www.cs.cmu.edu/~16385/ 16-385 Computer Vision Spring 2018, Lecture 10 Course announcements Homework 2 is due on February 23 rd. - Any questions about the homework? - How many of
More information